Yield curves from different bond data sets

It is well known that zero coupon rates are not observable variables. Their estimation process may be cumbersome and time consuming. We explore the extent to which the set of security prices used in the yield curve construction of three popular interest rate datasets (from the Federal Reserve Board, the US Department of the Treasury, and Bloomberg) may determine the results of different analyses. Using the same US Treasury prices from GovPX and applying the same fitting technique, we estimate zero coupon rates using different baskets of assets, i.e., including/excluding bills, on-the-run, and off-the-run bonds, attempting to mimic those used by each data providers. To illustrate the uncertainty surrounding these alternatives representations of the underlying yield curve, we examine common uses of these data sets in pricing, risk management and macroeconomic purposes. We find significant and sometime overwhelming differences in the volatility term structure, the pricing of interest rate derivatives, and the correlations among different forward rates particularly in both ends of the yield curve. Relevant implications are also observed on a classic test of the expectations hypothesis. The simplest asset basket, which only includes the on-the-run bills and bonds, is probably the one with the best results.